CN-121696767-B - Cutter state prediction method and system based on digital twin

CN121696767BCN 121696767 BCN121696767 BCN 121696767BCN-121696767-B

Abstract

The application relates to the technical field of equipment state monitoring and digital twin, and particularly discloses a cutter state prediction method and system based on digital twin. According to the method, whether the current cutter working condition data are required to be marked and classified and stored is judged according to the probability of applicability, the gate entropy value and the abrasion state parameter of the sensor feature vector and the processing condition vector of the expert network, model training is triggered according to the data caching condition in the classified cache region, the network model to be updated is determined according to the data caching condition, the model is not required to be completely retrained, retrained efficiency is improved, the network model to be updated is trained according to the cache data in the classified cache region, the processing capability of the model to various working condition data is improved, and further the cutter state prediction accuracy is improved.

Inventors

LI DONGPENG
LI WEIHUA
HE GUOLIN
JIANG JUNFENG
TANG JUN

Assignees

深圳市鼎粤科技有限公司
深圳华制智能制造技术有限公司

Dates

Publication Date: 20260508
Application Date: 20260214

Claims (10)

1. A digital twin-based tool state prediction method, which is characterized by being applied to a digital twin prediction system, wherein the digital twin prediction system comprises a physical entity layer, a connection and data layer, a twin model layer, a service and application layer and a synchronization and correction layer, and the method comprises the following steps: The connection and data layer acquires first cutter working condition data from the physical entity layer through a data acquisition interface, and processes the first cutter working condition data to obtain a sensor characteristic vector and a processing condition vector; The twin model layer processes the sensor feature vector and the processing condition vector to obtain the applicable probability, the gate entropy value and the first abrasion state vector of each expert network; the service and application layer performs state estimation based on the first wear state vector to obtain a wear state parameter; the synchronization and correction layer marks the first cutter working condition data when at least one of the applicable probability, the gating entropy value and the abrasion state parameter meets a preset condition, and stores the first cutter working condition data into a classification buffer area according to the label; The twin model layer determines a network model to be updated according to the data caching condition of the classified cache region, and trains the network model to be updated based on cache data in the classified cache region to obtain a latest twin model layer; The synchronization and correction layer synchronizes the latest twin model layer and the physical entity layer, so that the connection and data layer can acquire second tool working condition data from the physical entity layer through a data acquisition interface, and the second tool working condition data can be processed based on the latest twin model layer to acquire a second abrasion state vector.
2. The digital twinning-based tool state prediction method of claim 1, wherein the twinning model layer includes a gating network, a hybrid expert module including at least one expert network, and a physical information module including a twinning state prediction model.
3. The method for predicting a state of a tool based on digital twin according to claim 2, wherein the twin model layer determines a network model to be updated according to the data buffering condition of the classification buffer, trains the network model to be updated based on the buffered data in the classification buffer, and obtains a latest twin model layer, comprising: When the data caching condition is that the data volume of the brand new working condition data is larger than a first preset quantity threshold value, acquiring a pre-trained general expert network and the gating network as the network model to be updated; and adding the general expert network to the hybrid expert module, and training the general expert network and the gating network based on a preset learning rate dynamic adjustment strategy and brand new working condition data of the classification buffer region to obtain the latest twin model layer.
4. The method for predicting a state of a tool based on digital twinning according to claim 3, wherein the training the network model to be updated based on the cache data in the classified cache area to obtain a latest twinning model layer further comprises: Analyzing the brand new working condition data based on the gate control network to obtain the applicable probability of each expert network in the mixed expert module and the brand new working condition data, and respectively carrying out state prediction on the brand new working condition data based on each expert network to obtain a state prediction result; carrying out weighted average on each state prediction result based on the applicable probability to obtain a comprehensive prediction state; Acquiring a real abrasion state, and constructing a distillation loss function based on the real abrasion state and the comprehensive prediction state; And training the general expert network and the gating network based on the distillation loss function and the brand new working condition data to obtain the latest twin model layer.
5. The method for predicting a state of a tool based on digital twin according to claim 2, wherein the twin model layer determines a network model to be updated according to the data buffering condition of the classification buffer, trains the network model to be updated based on the buffered data in the classification buffer, and obtains a latest twin model layer, and further comprises: when the data caching condition is that the data quantity of the similar new working condition data is larger than a second preset quantity threshold value, acquiring an expert network corresponding to the similar working condition data as a reference expert network, wherein the similar new working condition data is new working condition data with similarity larger than a preset similarity threshold value with the similar working condition data; obtaining reference expert network parameters, generating a newly added expert network, and taking the gate control network and the newly added expert network as the network model to be updated; And adding the newly added expert network to the hybrid expert module, and carrying out fine adjustment on the gate control network and the fine-adjustable layer of the newly added expert network based on a preset difference constraint rule and the similar new working condition data to obtain the latest twin model layer.
6. The digital twin based tool state prediction method according to claim 1, wherein the twin model layer determines a network model to be updated according to the data buffering condition of the classification buffer, and further comprising: And when the data caching condition is that the data quantity of the known working condition variant data is larger than a third preset quantity threshold, taking the gating network as the network model to be updated.
7. The digital twin based tool state prediction method according to claim 2, wherein the processing the sensor feature vector and the machining condition vector by the twin model layer to obtain the probability of applicability, the gating entropy value, and the first wear state vector for each expert network comprises: processing the sensor feature vector and the processing condition vector based on the gating network to obtain the applicable probability and the gating entropy value of each expert network; When the difference value of the applicable probabilities is larger than or equal to a preset difference value threshold, processing the sensor feature vector and the processing condition vector based on the expert network with the highest applicable probability to obtain a twin state feature vector, wherein the difference value of the applicable probabilities is the difference value of the maximum applicable probability and the minimum applicable probability; and processing the sensor feature vector, the processing condition vector and the twin state feature vector based on the twin state prediction model to obtain a first wear state vector.
8. The method for predicting tool state based on digital twinning according to claim 7, wherein after the processing the sensor feature vector and the processing condition vector based on the gating network, obtaining the probability of applicability of each expert network and the gating entropy value, further comprises: when the applicable probability difference value is smaller than the preset difference value threshold value, processing the sensor feature vector and the processing condition vector based on each expert network to obtain each state feature sub-vector; Weighting and fusing all the state characteristic sub-vectors according to all the applicable probabilities to obtain the twin state characteristic vector; and processing the sensor feature vector, the processing condition vector and the twin state feature vector based on the twin state prediction model to obtain a first wear state vector.
9. The digital twin based tool state prediction method according to claim 2, wherein the twin model layer processes the sensor feature vector and the machining condition vector, and further comprises, before obtaining the probability of applicability, the gating entropy value, and the first wear state vector for each expert network: collecting a nominal working condition data set; Performing iterative training on a pre-training twin state prediction model based on a preset composite loss function and the nominal working condition data set to obtain gradient norms of loss items in the preset composite loss function; Acquiring a current training stage, and adjusting the weight coefficient of each loss term according to the current training stage, each gradient norm and a preset weight adjustment formula to acquire a composite loss function of the next iteration; And carrying out iterative training according to the composite loss function of the next iteration until reaching a preset condition, stopping iterative training, and obtaining the twin state prediction model.
10. The digital twinning-based tool state prediction method of claim 1, wherein the digital twinning prediction system further comprises a control feedback interface, wherein the service and application layer performs state estimation based on the first wear state vector, and wherein after obtaining the wear state parameter, further comprises: carrying out wear trend prediction on the historical wear data and the working condition change sequence based on the time sequence attention model to obtain a wear prediction result; Analyzing the wear state parameters and the wear prediction results based on a preset suggestion generation model to generate a tool adjustment suggestion; and the control feedback interface feeds back the tool adjustment advice to the physical entity layer.

Description

Cutter state prediction method and system based on digital twin Technical Field The application relates to the technical field of equipment state monitoring and digital twin, in particular to a cutter state prediction method and system based on digital twin. Background In automated manufacturing, accurate estimation of tool wear is critical to ensuring machining quality, preventing catastrophic tool failure, and optimizing tool life. The cutter works for a long time under severe working conditions such as high-speed cutting, intermittent cutting and the like, and degradation phenomena such as abrasion of a rear cutter face, crater abrasion and tipping and the like are extremely easy to occur. Deviations in tool health can lead to inaccurate machining geometry, poor surface finish, and can damage expensive workpieces, even causing safety risks and economic losses. Existing tool wear monitoring methods can be divided into two main categories, direct measurement and indirect inference. Direct measurement relies on offline means such as microscope or vision measurement, and the like, and the accuracy is high but the operation is stopped, so that the online monitoring requirement is difficult to meet. The indirect inference method infers the abrasion state by collecting sensor data such as vibration, force, torque, current and the like, and has the advantages of non-invasiveness and online application. The indirect inference method mostly adopts a physical-based model and a data-driven-based method. However, neither the physical-based model nor the data-driven method can cover various machining conditions, and the prediction accuracy drops suddenly when the conditions are switched. Therefore, how to improve the tool state prediction accuracy is a problem to be solved. Disclosure of Invention The application provides a cutter state prediction method and system based on digital twinning, which are used for improving cutter state prediction accuracy. In a first aspect, the present application provides a tool state prediction method based on digital twinning, which is applied to a digital twinning prediction system, wherein the digital twinning prediction system comprises a physical entity layer, a connection and data layer, a twinning model layer, a service and application layer and a synchronization and correction layer, and the method comprises: The connection and data layer acquires first cutter working condition data from the physical entity layer through a data acquisition interface, and processes the first cutter working condition data to obtain a sensor characteristic vector and a processing condition vector; The twin model layer processes the sensor feature vector and the processing condition vector to obtain the applicable probability, the gate entropy value and the first abrasion state vector of each expert network; the service and application layer performs state estimation based on the first wear state vector to obtain a wear state parameter; the synchronization and correction layer marks the cutter working condition data when at least one of the applicable probability, the gating entropy value and the abrasion state parameter meets preset conditions, and stores the cutter working condition data into a classification buffer area according to the label; The twin model layer determines a network model to be updated according to the data caching condition of the classified cache region, and trains the network model to be updated based on cache data in the classified cache region to obtain a latest twin model layer; The synchronization and correction layer synchronizes the latest twin model layer and the physical entity layer, so that the connection and data layer can acquire second tool working condition data from the physical entity layer through a data acquisition interface, and the second tool working condition data can be processed based on the latest twin model layer to acquire a second abrasion state vector. The application discloses a cutter state prediction method and a cutter state prediction system based on digital twinning, which judge whether the current cutter working condition data are required to be marked and stored in a classified mode according to the applicable probability, a gate entropy value and the abrasion state parameter of a sensor feature vector and a processing condition vector of an expert network, trigger model training according to the data caching condition in a classified cache region, determine a network model to be updated according to the data caching condition, and not need to completely retrain the model, thereby improving retrain efficiency, training the network model to be updated according to the cached data in the classified cache region, improving the processing capacity of the model to various working condition data, and further improving cutter state prediction accuracy. Drawings In order to more clearly illustrate the technical solutions of the embodiments of t